Abstract
Development of models of probabilistic phenomena such as low flows is often hindered by short record lengths. At-site estimates of parameters and low-flow quantiles may not be so reliable. In flood frequency analysis, one approach to overcome this difficulty is regional frequency analysis. The same general approach is applied here in the case of low-flow estimation. Daily streamflow data from 13 gauging stations in northwest Bangladesh are used. Weibull, also known as extreme type 3 for minimum, and log-Pearson type 3 distributions are found to be suitable for modelling low flows of the rivers in this region. The shape parameters of the distributions are estimated on a regional basis. Low-flow quantiles estimated by regional and at-site frequency analysis are compared.